110 lines
2.9 KiB
Go
110 lines
2.9 KiB
Go
package qwen
|
||
|
||
import (
|
||
"encoding/json"
|
||
"errors"
|
||
"fmt"
|
||
modelQwen "fonchain-fiee/pkg/model/qwen"
|
||
"fonchain-fiee/pkg/utils"
|
||
|
||
"go.uber.org/zap"
|
||
)
|
||
|
||
// VL 调用通义千问视觉多模态API，支持多个视频、多张图片和文本
|
||
func VL(videoURLs []string, imageURLs []string, text string, model string) (resp *modelQwen.VLResponse, err error) {
|
||
// 设置默认模型
|
||
if model == "" {
|
||
model = "qwen3-vl-plus"
|
||
}
|
||
|
||
// 构建内容列表
|
||
content := make([]modelQwen.VLContent, 0)
|
||
|
||
// 添加视频内容，支持自定义fps
|
||
for _, videoURL := range videoURLs {
|
||
fps := 2 // 默认fps为2
|
||
content = append(content, modelQwen.VLContent{
|
||
Type: "video_url",
|
||
VideoURL: &modelQwen.VideoURL{
|
||
URL: videoURL,
|
||
},
|
||
FPS: fps,
|
||
})
|
||
}
|
||
|
||
// 添加图片内容
|
||
for _, imageURL := range imageURLs {
|
||
content = append(content, modelQwen.VLContent{
|
||
Type: "image_url",
|
||
ImageURL: &modelQwen.ImageURL{
|
||
URL: imageURL,
|
||
},
|
||
})
|
||
}
|
||
|
||
// 添加文本内容
|
||
if text != "" {
|
||
content = append(content, modelQwen.VLContent{
|
||
Type: "text",
|
||
Text: text,
|
||
})
|
||
}
|
||
|
||
// 构建请求
|
||
req := modelQwen.VLRequest{
|
||
Model: model,
|
||
Messages: []modelQwen.VLMessage{
|
||
{
|
||
Role: "user",
|
||
Content: content,
|
||
},
|
||
},
|
||
}
|
||
|
||
// 序列化请求
|
||
jsonData, err := json.Marshal(req)
|
||
if err != nil {
|
||
zap.L().Error("VL Marshal failed", zap.Error(err))
|
||
return nil, errors.New("序列化请求失败")
|
||
}
|
||
|
||
// 发送请求，使用PostBytesHeader获取状态码和响应体
|
||
statusCode, body, err := utils.PostBytesHeader(modelQwen.DashscopeVLURL, map[string]interface{}{
|
||
"Authorization": "Bearer " + modelQwen.DashscopeAPIKey,
|
||
"Content-Type": "application/json",
|
||
// "X-DashScope-OssResourceResolve": "enable", // 启用OSS资源解析
|
||
}, jsonData)
|
||
if err != nil {
|
||
zap.L().Error("VL Post failed", zap.Error(err))
|
||
return nil, errors.New("请求视觉AI失败")
|
||
}
|
||
|
||
// 检查状态码，如果不是200，尝试解析错误响应
|
||
if statusCode != 200 {
|
||
// 尝试解析错误响应
|
||
var errorResp struct {
|
||
Error struct {
|
||
Message string `json:"message"`
|
||
Type string `json:"type"`
|
||
Code string `json:"code"`
|
||
} `json:"error"`
|
||
}
|
||
if err := json.Unmarshal(body, &errorResp); err == nil && errorResp.Error.Message != "" {
|
||
zap.L().Error("VL API error", zap.Int("status", statusCode), zap.String("message", errorResp.Error.Message))
|
||
return nil, fmt.Errorf("%s", errorResp.Error.Message)
|
||
}
|
||
// 如果无法解析错误响应，返回通用错误
|
||
zap.L().Error("VL API error", zap.Int("status", statusCode), zap.String("body", string(body)))
|
||
return nil, fmt.Errorf("接口返回错误")
|
||
}
|
||
|
||
// 解析响应
|
||
var result modelQwen.VLResponse
|
||
if err = json.Unmarshal(body, &result); err != nil {
|
||
zap.L().Error("VL Unmarshal failed", zap.Error(err), zap.String("body", string(body)))
|
||
return nil, fmt.Errorf("解析响应失败: %v", err)
|
||
}
|
||
|
||
return &result, nil
|
||
}
|